The Internet of Manufacturing Things - Drexel...
Transcript of The Internet of Manufacturing Things - Drexel...
The Internet of Manufacturing Things
Athulan Vijayaraghavan
CTO and Co-Founder, System Insights Berkeley, CA | Chennai, India
1
The Internet of Things
2
“interconnection of sensors, devices, and other things”
The Hype Cycle
3
Why Manufacturing?• Manufacturing is Big: $2
Trillion sector
• Discrete Manufacturing: Products for consumers and the supply chain
• High potential for productivity improvement
• Manufacturing generates a very large amount of data – most of it falls on the floor
4
Manufacturing Today
• Global
• Fragmented
• Heterogeneous
5
Improving the Transformation• Focus:
• Productivity? Profitability? Return on asset?
• Part quality? Employee safety?
• Sustainability? Energy usage?
• How about a more holistic view?
6
transformation
raw material
stuff
Track flow of resources and intelligence across manufacturing process
Process Improvement Design Integration Usage Analytics
7
Grand Challenge: Process Traceability
time
Foundry Forging Roughing Finishing Sub-Assembly WarehouseFinal
Assembly Shippingunique
part
Design Impacts
Usage ImpactsManufacturing
Enabling Technology
8
Standards Sensors Software
Internet of Things
Internet of Manufacturing Things
Standards
9
Why no Standards?
• Manufacturing data highly complex
• High barriers to entry
• Specialized technical knowledge deterrent to innovation
10
Why Standards!
11
What do you prefer?
MTConnect• Open royalty-free standard providing data from devices using a
common unambiguous vocabulary
• Uses XML and HTTP – Internet ready
• Simple, free, and extensible
• Design Goals
• Capture manufacturing domain model
• Read-only – inherently secure
• Borrow when you can
12
Differentiation• How is MTConnect different than OPC, Ethernet/IP?
• OPC is a transport – it gets data from one location to another
• Same with the field-buses
• OPC and Ethernet/IP do not have a domain model – nothing specific for manufacturing
• Tags are different for every implementation
13
Example: “Current Postion”• OPC
• Siemens Controller: /Channel/MachineAxis/actToolBasePos[u1,1] – returns a single value
• Bosch ControllerNC.Chan.AxisPosMcs,1 – returns a vector of values that needs to be decoded
• MTConnect
• Devices <Axes id=“a” name=“axes”> <Linear id="x1" name="X”> <DataItems> <DataItem category="SAMPLE" id="x2" name="Xact” subType="ACTUAL" type=“POSITION" units=“MILLIMETER"/> </DataItem></Linear></Axes>
• Stream of data<ComponentStream component="Linear"name="X" componentId="x1”><Samples>
<Position dataItemId=“x2” timestamp="2014-02-18T23:32:11.133634” name="Xact" sequence="2810505090” subType=“ACTUAL">2.0019900000</Position></Samples>
14
IoMT Requirements• Security
• MTConnect is read-only and inherently secure
• Reliability
• MTConnect provides a reliable messaging platform
• Performance
• MTConnect can run on embedded platforms
• One agent can support multiple devices
• Sensor data
• The most complete standardized sensor vocabulary and extensible for new sensors
• Support for realtime time-series data
15
IOT Standards• MQTT Protocol
• Lightweight IP-based M2M communication protocol
• Pub/sub model – highly scalable
• Large community, supported by Eclipse
• MQTT + MTConnect
• Take domain model of MTConnect and apply in MQTT
• MTConnect can sit below/above MQTT layer based on the application
16
The Digital Thread
17
Initial ModelIdea
Product
Design Refinement
Solid Model
Manufacturing Planning
Manufacturing Process
Metrology
Design
Manufacturing
Integrating manufacturing with the design phase
Standards along the way
18
Initial ModelIdea
Product
Design Refinement
Solid Model
Manufacturing Planning
Manufacturing Process
Metrology
Design
Manufacturing
STEP-NC MTConnect QIF/DMSC
STL, STEP, IGES
But can we do better?
• We have standards for the boxes – what about the arrows?
• How do we standardize analysis across the digital thread?
• We need cross-standard interoperability
19
Sensors
20
Sensors• Enable decision-making and automation
• Sensors at every level:
• Manufacturing process —> Supply Chain
• What we need:
• Minimally Invasive
• Physics based
• Open question: Where do you put the intelligence?
21
Sensor Intelligence• Traditional Approach: Self-contained local
command-and-control loops
• Is centralized intelligence possible?
• Challenges: data load, bandwidth, latency
• Solution:
• Distributed decentralized systems
• Split decisions between local and central controllers
22
centralized controllers
balance?
device controllers
Software
23
Software• Data Management:
• High data volumes
• Structured and unstructured data
• Decision-making:
• Event-based decision making
• Multi-dimensional reasoning
• Multiple temporal scales
24
Data Volumes
Small Shop: 2~10 TB/year Medium Shop: 5 ~ 25 TB/year Large Shop: 16 ~ 80 TB/year Enterprise: 80 ~ 5000 TB/year
US Machining Sector: 200 PB ~ 1XB/year
25
Data TypesStructured Unstructured Tribal Knowledge
Sensor Machine Telemetrics
Alarms, Faults Quality Control
Performance + Test
Annotations Over-rides
Interruptions
26
So what do we do with all of this data?
Event ReasoningEvent Processing
fusion
filter
aggregate
identify relationships
The Manufacturing Event Cloud
spindle speed position
alarms
notification
static datafeedrate overrides
Event: Something that happened at a point in time
tribal knowledge
27
Complex Event ProcessingTemporal
Overlap ContainsBefore/After
Spatial
Clustering TrendingShapes
Temporal Decision ScalesTemporal scales can vary from µ-seconds to days
anytime:process management
neartime:process improvement
realtime:process control
m-Seconds Seconds Hours Days
Process Interface
Sub-Components
Manufacturing Equipment
Manufacturing Supply Chain
Manufacturing Enterprise
Temporal Decision Scale
Man
ufac
turin
g A
naly
sis
Sca
le
28
Multidimensional Reasoning
:01 :02 :03 :04 :05 :06
V Mill 1
Lathe 1
Robot
CMM
Cell
Day
Line
Part 1Part 2
Part 6
......Batch
Power: 3562 WattsLoad[X]: 120%Alarm: ACTIVE
Time
Equi
pmen
t
Prod
uct
:01 :02 :03 :04 :05 :06
V Mill 1
Lathe 1
Robot
CMM
Cell
Day
Line
Part 1Part 2
Part 6
......Batch
Time
Equi
pmen
t
Prod
uct
Report: Total Energy UsagePeriod: DailyDevice: V-Mill 1
:01 :02 :03 :04 :05 :06
V Mill 1
Lathe 1
Robot
CMM
Cell
Day
Line
Part 1Part 2
Part 6
......Batch
Time
Equi
pmen
t
Prod
uct
Analysis: Power ConsumptionDevice: Cell 1Time: NowMulti-dimensional
reasoning allows us to slice data across any plane,
including: time, machine organization, parts
29
IoMT: Enabling Technology
30
High speed data from heterogeneous
sources
Integration across software and
hardware platforms
Decision-making across spatial and
temporal resolutions
IoMT vs. IoT
• Enterprise focused vs. Consumer Focused
• Islands of Excellence vs. New Ecosystems
• Mature markets vs. New Markets
• “the internet of what?” vs. “take my money!”
31
Closing Thoughts
• Terrific potential
• Being domain specific helps – a lot
• Don’t reinvent the wheel
• Boxes ok, Arrows need work
32